Hardware implementation of AI primitives

AI paradigms and techniques have, by and large, been developed under the influence of sequential computational models targeted at the von Neumann processor. This has, of necessity, determined the common languages and data structures employed in AI-related problem-solving. An exception is the cognitive modelling wing of AI which has consistently looked forward to a connectionist, or so-called parallel distributed processing, environment. But here again, AI researchers have usually been obliged to simulate their networks on sequential processors. It is time to look beyond and above the sequential computational model. Now that there exists the technology to implement affordable parallelism, designers of novel hardware should seek to identify AI primitives at the level of declarative and representational formalisms, rather than at the level of (sequential) programming languages and data structures. The potential gain in taking the higher view is that not only are execution speeds improved but software becomes less complex. >